Methods and devices for adaptive neural image compression, and computer-readable records

Abstract

The invention relates to a method, apparatus, and computer-readable recording device capable of adaptive neural image compression by supercomputation. The method may involve the creation of an alternative input image and an alternative target quality control parameter by means of an original input image and a target quality control parameter, wherein the alternative input image is a modified version of the original input image and the alternative target quality control parameter is a modified version of the target quality control parameter. The method may also include alternative input image encoding, based on the alternative input image and alternative target quality control parameters, using coding neural networks, to create a compressed representation of the alternative input image.